diff options
Diffstat (limited to 'meta-amd-bsp/recipes-kernel/linux/linux-yocto-4.14.71/1654-drm-amdgpu-Allow-get_user_pages-to-fail-in-restore-w.patch')
| -rw-r--r-- | meta-amd-bsp/recipes-kernel/linux/linux-yocto-4.14.71/1654-drm-amdgpu-Allow-get_user_pages-to-fail-in-restore-w.patch | 120 |
1 files changed, 120 insertions, 0 deletions
diff --git a/meta-amd-bsp/recipes-kernel/linux/linux-yocto-4.14.71/1654-drm-amdgpu-Allow-get_user_pages-to-fail-in-restore-w.patch b/meta-amd-bsp/recipes-kernel/linux/linux-yocto-4.14.71/1654-drm-amdgpu-Allow-get_user_pages-to-fail-in-restore-w.patch new file mode 100644 index 00000000..3dd1057d --- /dev/null +++ b/meta-amd-bsp/recipes-kernel/linux/linux-yocto-4.14.71/1654-drm-amdgpu-Allow-get_user_pages-to-fail-in-restore-w.patch @@ -0,0 +1,120 @@ +From 7d8e00c65331476edd5611b4ba9d161ac59000d7 Mon Sep 17 00:00:00 2001 +From: Felix Kuehling <Felix.Kuehling@amd.com> +Date: Wed, 5 Apr 2017 19:45:23 -0400 +Subject: [PATCH 1654/4131] drm/amdgpu: Allow get_user_pages to fail in restore + worker + +Avoid stalling queues indefinitely if get_user_pages fails. This type +of failure is indicative of a user mode bug where memory is freed that +is still mapped to the GPU. Let the restore continue and update the +page table with invalid entries for the failed BO. If the GPU tries to +access it, the application will crash with a VM fault. + +Bug: SWDEV-117987 +Change-Id: Ica23b8d562e8268d80109192f1c5f0c16eb72de0 +Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com> +--- + drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c | 52 ++++++++++++++++++------ + 1 file changed, 40 insertions(+), 12 deletions(-) + +diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c +index bde4f6a..3f1b1d9 100644 +--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c ++++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c +@@ -523,7 +523,7 @@ static int init_user_pages(struct kgd_mem *mem, struct mm_struct *mm, + ret = amdgpu_ttm_tt_get_user_pages(bo->tbo.ttm, mem->user_pages); + up_read(&mm->mmap_sem); + if (ret) { +- pr_err("%s: Failed to get user pages\n", __func__); ++ pr_err("%s: Failed to get user pages: %d\n", __func__, ret); + goto free_out; + } + +@@ -1996,8 +1996,14 @@ static int update_invalid_user_pages(struct amdkfd_process_info *process_info, + mem->user_pages); + if (ret) { + mem->user_pages[0] = NULL; +- pr_err("%s: Failed to get user pages\n", __func__); +- goto unlock_mmap_out; ++ pr_info("%s: Failed to get user pages: %d\n", ++ __func__, ret); ++ ret = 0; ++ /* Pretend it succeeded. It will fail later ++ * with a VM fault if the GPU tries to access ++ * it. Better than hanging indefinitely with ++ * stalled user mode queues. ++ */ + } + + /* Mark the BO as valid unless it was invalidated +@@ -2064,6 +2070,16 @@ static int validate_invalid_user_pages(struct amdkfd_process_info *process_info) + + amdgpu_sync_create(&sync); + ++ /* Avoid triggering eviction fences when unmapping invalid ++ * userptr BOs (waits for all fences, doesn't use ++ * FENCE_OWNER_VM) ++ */ ++ list_for_each_entry(peer_vm, &process_info->vm_list_head, ++ vm_list_node) ++ amdgpu_amdkfd_remove_eviction_fence(peer_vm->base.root.bo, ++ process_info->eviction_fence, ++ NULL, NULL); ++ + ret = process_validate_vms(process_info); + if (ret) + goto unreserve_out; +@@ -2076,15 +2092,17 @@ static int validate_invalid_user_pages(struct amdkfd_process_info *process_info) + + bo = mem->bo; + +- /* Copy pages array and validate the BO */ +- memcpy(bo->tbo.ttm->pages, mem->user_pages, +- sizeof(struct page *) * bo->tbo.ttm->num_pages); +- amdgpu_ttm_placement_from_domain(bo, mem->domain); +- ret = ttm_bo_validate(&bo->tbo, &bo->placement, +- false, false); +- if (ret) { +- pr_err("%s: failed to validate BO\n", __func__); +- goto unreserve_out; ++ /* Copy pages array and validate the BO if we got user pages */ ++ if (mem->user_pages[0]) { ++ memcpy(bo->tbo.ttm->pages, mem->user_pages, ++ sizeof(struct page *) * bo->tbo.ttm->num_pages); ++ amdgpu_ttm_placement_from_domain(bo, mem->domain); ++ ret = ttm_bo_validate(&bo->tbo, &bo->placement, ++ false, false); ++ if (ret) { ++ pr_err("%s: failed to validate BO\n", __func__); ++ goto unreserve_out; ++ } + } + + /* Validate succeeded, now the BO owns the pages, free +@@ -2097,6 +2115,12 @@ static int validate_invalid_user_pages(struct amdkfd_process_info *process_info) + list_move_tail(&mem->validate_list.head, + &process_info->userptr_valid_list); + ++ /* Update mapping. If the BO was not validated ++ * (because we couldn't get user pages), this will ++ * clear the page table entries, which will result in ++ * VM faults if the GPU tries to access the invalid ++ * memory. ++ */ + list_for_each_entry(bo_va_entry, &mem->bo_va_list, bo_list) { + if (!bo_va_entry->is_mapped) + continue; +@@ -2113,6 +2137,10 @@ static int validate_invalid_user_pages(struct amdkfd_process_info *process_info) + } + } + unreserve_out: ++ list_for_each_entry(peer_vm, &process_info->vm_list_head, ++ vm_list_node) ++ amdgpu_bo_fence(peer_vm->base.root.bo, ++ &process_info->eviction_fence->base, true); + ttm_eu_backoff_reservation(&ticket, &resv_list); + amdgpu_sync_wait(&sync); + amdgpu_sync_free(&sync); +-- +2.7.4 + |